METHOD GUIDE
SEM with SPSS AMOS
Preparing the SPSS Data and Reading the Fit Indices
SEM in SPSS has one catch: SPSS Statistics does not fit structural equation models. AMOS, a separate program, does, and it reads the data file you prepare in SPSS. Preparing that file, reading the fit table AMOS prints and wording the result for a reviewer are the three tasks that follow.
IN SHORT
What SEM Estimates and Where SPSS Stops
Structural equation modeling joins a measurement model, which links observed items to latent constructs, with a structural model of the relationships among those constructs, and it asks whether the specified structure is consistent with the covariances in the data. A good fit shows consistency, and not that the theory is correct. To carry out structural equation modeling in SPSS, which in practice means AMOS, screen and save the data in SPSS Statistics, open the file in AMOS, draw or specify the measurement and structural paths, choose the estimation settings and the output, and read the fit table before the estimates (labels vary by AMOS version). OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, carries out and reviews this analysis through an in-house PhD-level statistician.
ANALYSIS SUPPORT
Analysis Support for This Method
A structural model is only as sound as its measurement model, and AMOS describes the support available for the program itself.
Structural Equation Modeling (SEM)
The structural model specified, fitted and judged within your study design, with the measurement model, the fit evidence and the reporting agreed for your data.
Confirmatory Factor Analysis (CFA)
The measurement model on its own, for when the factor structure of the scales has to be established before relationships among constructs are tested.
BEFORE YOU RUN IT
Getting the SPSS File Ready for AMOS
AMOS works from what the SPSS file holds, so most of what stops a model from running, or makes it untrustworthy, can be found in SPSS first.
Coding and Variable Names SPSS data for AMOS is easier to work with when items are scored in one direction, with short unique names and clear labels. Reverse-keyed items are recoded before modeling begins.
Missing Values Listwise deletion shrinks the sample. AMOS can use all available data when means and intercepts are estimated, though some options are then unavailable, so the pattern of missingness is examined first.
Distribution of the Items Basic maximum likelihood assumes multivariate normality. Skewness, kurtosis and the number of response categories are checked, with a plan for marked departures such as resampling-based standard errors.
Sample Size and Model Size No single rule fixes the sample size. Needs rise with the number of parameters and fall with strong loadings, so the cases available are related to the model's complexity and any limit is stated.
IN SPSS
How to Run SEM in SPSS AMOS
AMOS is a separate program from IBM SPSS, which prepares the data it reads, and the outline below varies with the AMOS version and the estimation settings.
- In SPSS Statistics, screen the variables for coding errors, out-of-range values and missing data, then save the cleaned file as an SPSS data file.
- Open AMOS and select that file as the data source for the model.
- Draw the measurement model, with each indicator loading on its latent variable, then the structural paths among the latent variables.
- Name each observed variable as it appears in the data file, and set the scale of each latent variable by fixing one loading to 1 or its variance.
- In the analysis properties, choose the estimation method, standardized estimates and the fit measures to print.
- Run the model and check the notes and warnings before reading any fit index.
READING THE OUTPUT
Which AMOS Tables Carry the Fit Evidence
AMOS model fit indices appear in one Model Fit table with many rows, and no single row settles whether the model is acceptable.
| Output element | What it tells you | What to check |
|---|---|---|
| Notes for Model | The chi-square (CMIN), degrees of freedom and probability level for exact fit | That a significant chi-square is reported and not skipped, even though it is sensitive to sample size |
| CMIN/DF | The chi-square divided by its degrees of freedom | That it is treated as a descriptive ratio, since no universal cutoff exists |
| RMSEA with LO 90, HI 90 and PCLOSE | The error of approximation per degree of freedom, its 90% interval and a test of close fit | That the interval is reported, not only the point value |
| CFI and TLI in the Baseline Comparisons table | Fit relative to the independence model | That cutoffs are cited as guidance from simulation studies and not as pass-or-fail rules |
| RMR (standardized version through a plugin) | The average residual covariance, in the units of the data | That SRMR, which AMOS does not print in the Model Fit table, is obtained from the Standardized RMR plugin |
| Standardized Regression Weights and Squared Multiple Correlations | Loadings and structural paths on a common scale, and variance explained in each endogenous variable | That a loading above 1 or a negative variance is investigated as an improper solution |
| Modification Indices | The expected drop in chi-square if one fixed parameter were freed | That any change has a substantive reason and that the revised model is reported as exploratory |
REPORTING
Reporting Fit Honestly, Then the Estimates
Fit is reported as a set of indices with the model description in front of them, and every blank comes from your own AMOS output.
Describe the model before its fit: the constructs, the indicators of each, the data source, the estimation method and how missing data were handled. Then report χ²(df = ___) = ___, p = ___; RMSEA = ___, 90% CI [___, ___]; CFI = ___; TLI = ___; and SRMR = ___ where it was obtained. State which cutoffs served as guidance and why. Present the standardized loadings and structural paths with their standard errors or intervals, and note equivalent models that would fit the same covariances equally well. Because the fit indices show consistency with the data, the wording is that the model is consistent with the data, and not that the theory is confirmed. Identify the AMOS version in the methods text.
COMMON MISTAKES
Fit Reports That Draw Reviewer Objections
Reviewers of SEM papers tend to raise the same few concerns, and most can be seen in the data file or the output before submission.
Only the Indices That Passed Leaving out the chi-square or a weaker index hides disagreement among them. A set of indices is chosen in advance and reported in full.
Modification Until Fit Freeing parameters one at a time until the fit is acceptable tailors the model to the sample. Each change needs a substantive reason and the result is labeled exploratory.
Paths Fitted Before the Measurement Model Fitting the full structural model before the measurement model has been tested leaves poor fit hard to locate, in the constructs or in the paths between them. The measurement model is assessed first, and the structural paths are added once it holds.
Heywood Cases Passed Over A negative error variance or a standardized loading above 1 marks an improper solution. It points to misspecification, a small sample or too few indicators, and it is investigated before any index is quoted.
Indirect Paths Without Intervals An indirect path reported as the product of two estimates, with no interval from resampling or another method, leaves its uncertainty unstated.
Constructs Named Before Checked Giving a latent variable a construct's name does not show that the construct exists. The label comes from the items that load, and the text states what they measure.
Fit table and path diagram finished? Independent Statistical Review reads the specification, the fit evidence and the standardized estimates before you submit.
FREQUENTLY ASKED QUESTIONS
Questions About SEM and Fit Output in AMOS
The step from SPSS Statistics to AMOS raises practical questions before any statistical one, starting with whether AMOS is needed at all.
Can I run SEM in SPSS Statistics without AMOS?
No: SPSS Statistics itself does not fit structural equation models, so the model is fitted in AMOS or in another SEM program. A path model among observed variables can be approximated by a series of regressions in SPSS, but that gives no latent variables and no overall test of fit. Programs such as R, Mplus and Stata are used for the same models, and the choice follows your estimator, your data and your reporting needs.
What should the SPSS data file contain before I open AMOS?
Only the variables the model needs, scored in one direction, with missing values handled in a way you can describe. Work on a cleaned copy and keep the original file untouched. AMOS can also start from a covariance matrix with the sample size, but missing-data estimation and bootstrapping each need the raw cases, and AMOS does not bootstrap in a run that estimates missing data, so the SPSS data for AMOS are normally the raw cases with the missing-data plan settled first.
Where is SRMR in the AMOS output?
It is not in the Model Fit table, which prints RMR; the standardized version comes from the Standardized RMR plugin in the Plugins menu. The plugin window is opened and the model is then calculated again before a value appears. Report SRMR with the other indices, and name the AMOS version used.
Can OnlineSPSS prepare my SPSS data and run the model in AMOS?
Yes. An in-house PhD-level statistician can screen the file, specify the measurement and structural models, run them in AMOS and judge the fit evidence, or review a model you have already fitted. Presenting a settled model, from the fit table to the standardized paths, falls under Results Reporting & Interpretation. The constructs and the theory linking them remain yours, as do the conclusions.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as the dataset or the output you have, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Fit Indices Pulling
in Different Directions?
When the chi-square, RMSEA and CFI disagree, the specification, the sample and the data file often explain why, and a statistician can look for the cause before any parameter is freed.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014